Choosing the Dependent Variable in SAR Backscatter – Forest Biomass Models

Synthetic aperture radar (SAR) is a promising technology for mapping forest characteristics, and the anticipated open-access, L-band data from the NISAR mission holds potential for rapid and near-continuous monitoring. A dominant method for assessing forest characteristics using SAR backscatter is fitting and inversion of simple, semi-physical models. Models such as the Simple Water Cloud Model (SWCM) can be inverted mathematically; however, it is not necessarily true that the inverted model provides reliable predictions of forest characteristics, even when backscatter falls between the limits predicted by the original model. The challenge is inherently statistical, and not entirely dependent on model formulation. We illustrate the problem using forest aboveground biomass and L-band backscatter from a previous study in New Hampshire and Maine, USA. We suggest a greater focus on the role and purpose of regression models can help alleviate misconceptions and potential misapplications, leading to more informed and reliable prediction.